Jay Unick posted on Friday, March 10, 2006 - 6:03 pm
I have a question regarding LCA and complex sampling designs. I am using the National Comorbidity Survey that includes strata, psu’s and weights. But, I am only using a sub-sample of that survey not the whole population for my analysis, however, all the PSUs and Strata are present. I have been warned that when using a program such as Stata I need to have the whole population present but use a command subpop to subset my data in order to obtain accurate estimates. My question is, do I need to let Mplus 3.13 know that I am using a subpopulation and if so how can I do that?
You need Mplus Version 4 to handle subpopulations with complex survey data - see the Version 4 User's Guide on our web site, page 403, which includes a reference to Korn & Graubard's book.
anonymous posted on Wednesday, March 31, 2010 - 8:38 am
Hi, I have used the subpopulation command so that only a subgroup of participants are analyzed in an LCA. However, when I use the subpopulation command in conjunction with assigning participants to classes(using the savedata command), all participants are assigned, even though they were not included in the LCA. Is this correct? thanks!
This is correct. The way subpopulation is implemented is that we set to 0 the sampling weights for people that are not in the subpopulation. With zero weight they do not contribute to the estimation, however, the SAVEDATA command which is simply a data saving tool will include the people that are not in the subpopulation.
I'm using national survey data. When I save the assigned classes by using SAVEDATA command, it seems that the number of people in each class doesn't match what I got from "CLASSIFICATION OF INDIVIDUALS BASED ON THEIR MOST LIKELY LATENT CLASS MEMBERSHIP" in the output. I also gave it a try, the number matches when I took away settings of survey design.
How do I save the correct assigned classes given survey design? Thanks!
When a weight variable is used, the saved posterior probabilities are not weighted. I believe this is the difference you are seeing. If not, please send your files and license number to email@example.com.
Then, how do we determine the assigned classes for each participant given weight? I'm not quite sure how to calculate weighted posterior probabilities for each person. Could you please provide the formula or reference that I can use? Thank you.